2 * QEMU Host Memory Backend
4 * Copyright (C) 2013-2014 Red Hat Inc
7 * Igor Mammedov <imammedo@redhat.com>
9 * This work is licensed under the terms of the GNU GPL, version 2 or later.
10 * See the COPYING file in the top-level directory.
13 #include "qemu/osdep.h"
14 #include "sysemu/hostmem.h"
15 #include "hw/boards.h"
16 #include "qapi/error.h"
17 #include "qapi/qapi-builtin-visit.h"
18 #include "qapi/visitor.h"
19 #include "qemu/config-file.h"
20 #include "qom/object_interfaces.h"
21 #include "qemu/mmap-alloc.h"
22 #include "qemu/madvise.h"
27 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_DEFAULT
!= MPOL_DEFAULT
);
29 * HOST_MEM_POLICY_PREFERRED may either translate to MPOL_PREFERRED or
30 * MPOL_PREFERRED_MANY, see comments further below.
32 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_PREFERRED
!= MPOL_PREFERRED
);
33 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_BIND
!= MPOL_BIND
);
34 QEMU_BUILD_BUG_ON(HOST_MEM_POLICY_INTERLEAVE
!= MPOL_INTERLEAVE
);
38 host_memory_backend_get_name(HostMemoryBackend
*backend
)
40 if (!backend
->use_canonical_path
) {
41 return g_strdup(object_get_canonical_path_component(OBJECT(backend
)));
44 return object_get_canonical_path(OBJECT(backend
));
48 host_memory_backend_get_size(Object
*obj
, Visitor
*v
, const char *name
,
49 void *opaque
, Error
**errp
)
51 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
52 uint64_t value
= backend
->size
;
54 visit_type_size(v
, name
, &value
, errp
);
58 host_memory_backend_set_size(Object
*obj
, Visitor
*v
, const char *name
,
59 void *opaque
, Error
**errp
)
61 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
64 if (host_memory_backend_mr_inited(backend
)) {
65 error_setg(errp
, "cannot change property %s of %s ", name
,
66 object_get_typename(obj
));
70 if (!visit_type_size(v
, name
, &value
, errp
)) {
75 "property '%s' of %s doesn't take value '%" PRIu64
"'",
76 name
, object_get_typename(obj
), value
);
79 backend
->size
= value
;
83 host_memory_backend_get_host_nodes(Object
*obj
, Visitor
*v
, const char *name
,
84 void *opaque
, Error
**errp
)
86 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
87 uint16List
*host_nodes
= NULL
;
88 uint16List
**tail
= &host_nodes
;
91 value
= find_first_bit(backend
->host_nodes
, MAX_NODES
);
92 if (value
== MAX_NODES
) {
96 QAPI_LIST_APPEND(tail
, value
);
99 value
= find_next_bit(backend
->host_nodes
, MAX_NODES
, value
+ 1);
100 if (value
== MAX_NODES
) {
104 QAPI_LIST_APPEND(tail
, value
);
108 visit_type_uint16List(v
, name
, &host_nodes
, errp
);
109 qapi_free_uint16List(host_nodes
);
113 host_memory_backend_set_host_nodes(Object
*obj
, Visitor
*v
, const char *name
,
114 void *opaque
, Error
**errp
)
117 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
118 uint16List
*l
, *host_nodes
= NULL
;
120 visit_type_uint16List(v
, name
, &host_nodes
, errp
);
122 for (l
= host_nodes
; l
; l
= l
->next
) {
123 if (l
->value
>= MAX_NODES
) {
124 error_setg(errp
, "Invalid host-nodes value: %d", l
->value
);
129 for (l
= host_nodes
; l
; l
= l
->next
) {
130 bitmap_set(backend
->host_nodes
, l
->value
, 1);
134 qapi_free_uint16List(host_nodes
);
136 error_setg(errp
, "NUMA node binding are not supported by this QEMU");
141 host_memory_backend_get_policy(Object
*obj
, Error
**errp G_GNUC_UNUSED
)
143 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
144 return backend
->policy
;
148 host_memory_backend_set_policy(Object
*obj
, int policy
, Error
**errp
)
150 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
151 backend
->policy
= policy
;
154 if (policy
!= HOST_MEM_POLICY_DEFAULT
) {
155 error_setg(errp
, "NUMA policies are not supported by this QEMU");
160 static bool host_memory_backend_get_merge(Object
*obj
, Error
**errp
)
162 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
164 return backend
->merge
;
167 static void host_memory_backend_set_merge(Object
*obj
, bool value
, Error
**errp
)
169 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
171 if (!host_memory_backend_mr_inited(backend
)) {
172 backend
->merge
= value
;
176 if (value
!= backend
->merge
) {
177 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
178 uint64_t sz
= memory_region_size(&backend
->mr
);
180 qemu_madvise(ptr
, sz
,
181 value
? QEMU_MADV_MERGEABLE
: QEMU_MADV_UNMERGEABLE
);
182 backend
->merge
= value
;
186 static bool host_memory_backend_get_dump(Object
*obj
, Error
**errp
)
188 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
190 return backend
->dump
;
193 static void host_memory_backend_set_dump(Object
*obj
, bool value
, Error
**errp
)
195 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
197 if (!host_memory_backend_mr_inited(backend
)) {
198 backend
->dump
= value
;
202 if (value
!= backend
->dump
) {
203 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
204 uint64_t sz
= memory_region_size(&backend
->mr
);
206 qemu_madvise(ptr
, sz
,
207 value
? QEMU_MADV_DODUMP
: QEMU_MADV_DONTDUMP
);
208 backend
->dump
= value
;
212 static bool host_memory_backend_get_prealloc(Object
*obj
, Error
**errp
)
214 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
216 return backend
->prealloc
;
219 static void host_memory_backend_set_prealloc(Object
*obj
, bool value
,
222 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
224 if (!backend
->reserve
&& value
) {
225 error_setg(errp
, "'prealloc=on' and 'reserve=off' are incompatible");
229 if (!host_memory_backend_mr_inited(backend
)) {
230 backend
->prealloc
= value
;
234 if (value
&& !backend
->prealloc
) {
235 int fd
= memory_region_get_fd(&backend
->mr
);
236 void *ptr
= memory_region_get_ram_ptr(&backend
->mr
);
237 uint64_t sz
= memory_region_size(&backend
->mr
);
239 if (!qemu_prealloc_mem(fd
, ptr
, sz
, backend
->prealloc_threads
,
240 backend
->prealloc_context
, errp
)) {
243 backend
->prealloc
= true;
247 static void host_memory_backend_get_prealloc_threads(Object
*obj
, Visitor
*v
,
248 const char *name
, void *opaque
, Error
**errp
)
250 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
251 visit_type_uint32(v
, name
, &backend
->prealloc_threads
, errp
);
254 static void host_memory_backend_set_prealloc_threads(Object
*obj
, Visitor
*v
,
255 const char *name
, void *opaque
, Error
**errp
)
257 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
260 if (!visit_type_uint32(v
, name
, &value
, errp
)) {
264 error_setg(errp
, "property '%s' of %s doesn't take value '%d'", name
,
265 object_get_typename(obj
), value
);
268 backend
->prealloc_threads
= value
;
271 static void host_memory_backend_init(Object
*obj
)
273 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
274 MachineState
*machine
= MACHINE(qdev_get_machine());
276 /* TODO: convert access to globals to compat properties */
277 backend
->merge
= machine_mem_merge(machine
);
278 backend
->dump
= machine_dump_guest_core(machine
);
279 backend
->reserve
= true;
280 backend
->prealloc_threads
= machine
->smp
.cpus
;
283 static void host_memory_backend_post_init(Object
*obj
)
285 object_apply_compat_props(obj
);
288 bool host_memory_backend_mr_inited(HostMemoryBackend
*backend
)
291 * NOTE: We forbid zero-length memory backend, so here zero means
292 * "we haven't inited the backend memory region yet".
294 return memory_region_size(&backend
->mr
) != 0;
297 MemoryRegion
*host_memory_backend_get_memory(HostMemoryBackend
*backend
)
299 return host_memory_backend_mr_inited(backend
) ? &backend
->mr
: NULL
;
302 void host_memory_backend_set_mapped(HostMemoryBackend
*backend
, bool mapped
)
304 backend
->is_mapped
= mapped
;
307 bool host_memory_backend_is_mapped(HostMemoryBackend
*backend
)
309 return backend
->is_mapped
;
312 size_t host_memory_backend_pagesize(HostMemoryBackend
*memdev
)
314 size_t pagesize
= qemu_ram_pagesize(memdev
->mr
.ram_block
);
315 g_assert(pagesize
>= qemu_real_host_page_size());
320 host_memory_backend_memory_complete(UserCreatable
*uc
, Error
**errp
)
322 HostMemoryBackend
*backend
= MEMORY_BACKEND(uc
);
323 HostMemoryBackendClass
*bc
= MEMORY_BACKEND_GET_CLASS(uc
);
330 if (!bc
->alloc(backend
, errp
)) {
334 ptr
= memory_region_get_ram_ptr(&backend
->mr
);
335 sz
= memory_region_size(&backend
->mr
);
337 if (backend
->merge
) {
338 qemu_madvise(ptr
, sz
, QEMU_MADV_MERGEABLE
);
340 if (!backend
->dump
) {
341 qemu_madvise(ptr
, sz
, QEMU_MADV_DONTDUMP
);
344 unsigned long lastbit
= find_last_bit(backend
->host_nodes
, MAX_NODES
);
345 /* lastbit == MAX_NODES means maxnode = 0 */
346 unsigned long maxnode
= (lastbit
+ 1) % (MAX_NODES
+ 1);
347 /* ensure policy won't be ignored in case memory is preallocated
348 * before mbind(). note: MPOL_MF_STRICT is ignored on hugepages so
349 * this doesn't catch hugepage case. */
350 unsigned flags
= MPOL_MF_STRICT
| MPOL_MF_MOVE
;
351 int mode
= backend
->policy
;
353 /* check for invalid host-nodes and policies and give more verbose
354 * error messages than mbind(). */
355 if (maxnode
&& backend
->policy
== MPOL_DEFAULT
) {
356 error_setg(errp
, "host-nodes must be empty for policy default,"
357 " or you should explicitly specify a policy other"
360 } else if (maxnode
== 0 && backend
->policy
!= MPOL_DEFAULT
) {
361 error_setg(errp
, "host-nodes must be set for policy %s",
362 HostMemPolicy_str(backend
->policy
));
366 /* We can have up to MAX_NODES nodes, but we need to pass maxnode+1
367 * as argument to mbind() due to an old Linux bug (feature?) which
368 * cuts off the last specified node. This means backend->host_nodes
369 * must have MAX_NODES+1 bits available.
371 assert(sizeof(backend
->host_nodes
) >=
372 BITS_TO_LONGS(MAX_NODES
+ 1) * sizeof(unsigned long));
373 assert(maxnode
<= MAX_NODES
);
375 #ifdef HAVE_NUMA_HAS_PREFERRED_MANY
376 if (mode
== MPOL_PREFERRED
&& numa_has_preferred_many() > 0) {
378 * Replace with MPOL_PREFERRED_MANY otherwise the mbind() below
379 * silently picks the first node.
381 mode
= MPOL_PREFERRED_MANY
;
386 mbind(ptr
, sz
, mode
, backend
->host_nodes
, maxnode
+ 1, flags
)) {
387 if (backend
->policy
!= MPOL_DEFAULT
|| errno
!= ENOSYS
) {
388 error_setg_errno(errp
, errno
,
389 "cannot bind memory to host NUMA nodes");
394 /* Preallocate memory after the NUMA policy has been instantiated.
395 * This is necessary to guarantee memory is allocated with
396 * specified NUMA policy in place.
398 if (backend
->prealloc
&& !qemu_prealloc_mem(memory_region_get_fd(&backend
->mr
),
400 backend
->prealloc_threads
,
401 backend
->prealloc_context
, errp
)) {
407 host_memory_backend_can_be_deleted(UserCreatable
*uc
)
409 if (host_memory_backend_is_mapped(MEMORY_BACKEND(uc
))) {
416 static bool host_memory_backend_get_share(Object
*o
, Error
**errp
)
418 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
420 return backend
->share
;
423 static void host_memory_backend_set_share(Object
*o
, bool value
, Error
**errp
)
425 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
427 if (host_memory_backend_mr_inited(backend
)) {
428 error_setg(errp
, "cannot change property value");
431 backend
->share
= value
;
435 static bool host_memory_backend_get_reserve(Object
*o
, Error
**errp
)
437 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
439 return backend
->reserve
;
442 static void host_memory_backend_set_reserve(Object
*o
, bool value
, Error
**errp
)
444 HostMemoryBackend
*backend
= MEMORY_BACKEND(o
);
446 if (host_memory_backend_mr_inited(backend
)) {
447 error_setg(errp
, "cannot change property value");
450 if (backend
->prealloc
&& !value
) {
451 error_setg(errp
, "'prealloc=on' and 'reserve=off' are incompatible");
454 backend
->reserve
= value
;
456 #endif /* CONFIG_LINUX */
459 host_memory_backend_get_use_canonical_path(Object
*obj
, Error
**errp
)
461 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
463 return backend
->use_canonical_path
;
467 host_memory_backend_set_use_canonical_path(Object
*obj
, bool value
,
470 HostMemoryBackend
*backend
= MEMORY_BACKEND(obj
);
472 backend
->use_canonical_path
= value
;
476 host_memory_backend_class_init(ObjectClass
*oc
, void *data
)
478 UserCreatableClass
*ucc
= USER_CREATABLE_CLASS(oc
);
480 ucc
->complete
= host_memory_backend_memory_complete
;
481 ucc
->can_be_deleted
= host_memory_backend_can_be_deleted
;
483 object_class_property_add_bool(oc
, "merge",
484 host_memory_backend_get_merge
,
485 host_memory_backend_set_merge
);
486 object_class_property_set_description(oc
, "merge",
487 "Mark memory as mergeable");
488 object_class_property_add_bool(oc
, "dump",
489 host_memory_backend_get_dump
,
490 host_memory_backend_set_dump
);
491 object_class_property_set_description(oc
, "dump",
492 "Set to 'off' to exclude from core dump");
493 object_class_property_add_bool(oc
, "prealloc",
494 host_memory_backend_get_prealloc
,
495 host_memory_backend_set_prealloc
);
496 object_class_property_set_description(oc
, "prealloc",
497 "Preallocate memory");
498 object_class_property_add(oc
, "prealloc-threads", "int",
499 host_memory_backend_get_prealloc_threads
,
500 host_memory_backend_set_prealloc_threads
,
502 object_class_property_set_description(oc
, "prealloc-threads",
503 "Number of CPU threads to use for prealloc");
504 object_class_property_add_link(oc
, "prealloc-context",
505 TYPE_THREAD_CONTEXT
, offsetof(HostMemoryBackend
, prealloc_context
),
506 object_property_allow_set_link
, OBJ_PROP_LINK_STRONG
);
507 object_class_property_set_description(oc
, "prealloc-context",
508 "Context to use for creating CPU threads for preallocation");
509 object_class_property_add(oc
, "size", "int",
510 host_memory_backend_get_size
,
511 host_memory_backend_set_size
,
513 object_class_property_set_description(oc
, "size",
514 "Size of the memory region (ex: 500M)");
515 object_class_property_add(oc
, "host-nodes", "int",
516 host_memory_backend_get_host_nodes
,
517 host_memory_backend_set_host_nodes
,
519 object_class_property_set_description(oc
, "host-nodes",
520 "Binds memory to the list of NUMA host nodes");
521 object_class_property_add_enum(oc
, "policy", "HostMemPolicy",
522 &HostMemPolicy_lookup
,
523 host_memory_backend_get_policy
,
524 host_memory_backend_set_policy
);
525 object_class_property_set_description(oc
, "policy",
526 "Set the NUMA policy");
527 object_class_property_add_bool(oc
, "share",
528 host_memory_backend_get_share
, host_memory_backend_set_share
);
529 object_class_property_set_description(oc
, "share",
530 "Mark the memory as private to QEMU or shared");
532 object_class_property_add_bool(oc
, "reserve",
533 host_memory_backend_get_reserve
, host_memory_backend_set_reserve
);
534 object_class_property_set_description(oc
, "reserve",
535 "Reserve swap space (or huge pages) if applicable");
536 #endif /* CONFIG_LINUX */
538 * Do not delete/rename option. This option must be considered stable
539 * (as if it didn't have the 'x-' prefix including deprecation period) as
540 * long as 4.0 and older machine types exists.
541 * Option will be used by upper layers to override (disable) canonical path
542 * for ramblock-id set by compat properties on old machine types ( <= 4.0),
543 * to keep migration working when backend is used for main RAM with
544 * -machine memory-backend= option (main RAM historically used prefix-less
547 object_class_property_add_bool(oc
, "x-use-canonical-path-for-ramblock-id",
548 host_memory_backend_get_use_canonical_path
,
549 host_memory_backend_set_use_canonical_path
);
552 static const TypeInfo host_memory_backend_info
= {
553 .name
= TYPE_MEMORY_BACKEND
,
554 .parent
= TYPE_OBJECT
,
556 .class_size
= sizeof(HostMemoryBackendClass
),
557 .class_init
= host_memory_backend_class_init
,
558 .instance_size
= sizeof(HostMemoryBackend
),
559 .instance_init
= host_memory_backend_init
,
560 .instance_post_init
= host_memory_backend_post_init
,
561 .interfaces
= (InterfaceInfo
[]) {
562 { TYPE_USER_CREATABLE
},
567 static void register_types(void)
569 type_register_static(&host_memory_backend_info
);
572 type_init(register_types
);